Svmuu News: Xiaomi miclaw, an AI interaction testing product built on Xiaomi’s MiMo large language model, begins a small-scale closed beta today. Participation is by invitation only; no public recruitment is being conducted. Xiaomi miclaw runs as a system application, integrating over 50 system-level tools and ecosystem services. It employs an inference-execution loop engine, allowing the model to autonomously determine the sequence and parameters of tool invocations. The product focuses on validating the large model’s execution capabilities within Xiaomi’s “human-vehicle-home full ecosystem” and features a four-tier capability architecture: system-level capabilities, personal context understanding, ecosystem interconnectivity, and self-evolution. In terms of ecosystem interconnection, Xiaomi miclaw implements a complete Mi Home protocol client, enabling it to control IoT devices connected to Mi Home with user authorization, and supports the MCP protocol and third-party application integration via SDK. In terms of self-evolution, the product possesses four core capabilities: file-level memory, creation of sub-agents, MCP service configuration, and sandbox script execution, enabling it to autonomously build a memory system and expand its tool set. Xiaomi notes that the product is still undergoing optimization in terms of stability, power consumption, and execution success rates in complex scenarios. It recommends that users back up their data in advance and test the product in a controlled environment.